Artificial Intelligence Sensitivity Analysis of Community Clinic Stillbirth Risk Inference Under Unmeasured Gestational Age Error from Last-Menstrual-Period Recall: A Healthcare Probabilistic Bias Evaluation

Gestational age error from last-menstrual-period recall biases stillbirth risk inference in community clinics, yet the sensitivity of artificial intelligence estimators to this unmeasured error remains unquantified. the present analysis evaluates the finite-sample behaviour of an ordinary least squares estimator for a stillbirth risk model in which gestational age is measured with recall error. A replication-based design was used: 600 data sets of N = 500 were drawn from a declared data-generating process with intercept 12.0, exposure 0.42, group -3.1, severity 1.35, and error SD 4.2. The estimator was applied to each replication, and bias, root mean squared error, and confidence interval coverage were computed. The estimated bias for the intercept was 0.04257 and for the exposure coefficient -0.00076; the root mean squared error was 0.90516 for the intercept and 0.01633 for the exposure. Empirical coverage of the nominal 95% confidence intervals was 1.0 for both parameters. The sampling distribution of the exposure coefficient was approximately normal, and the Q-Q plot showed no material departure from normality. A receiver operating characteristic curve was constructed to assess classification performance. The results indicate that under the stated data-generating process, the ordinary least squares estimator is essentially unbiased for the exposure coefficient and achieves conservative coverage, but the intercept exhibits small positive bias. These findings provide a probabilistic benchmark for sensitivity analysis of stillbirth risk inference when gestational age is measured with error, and they suggest that artificial intelligence systems trained on such data should account for intercept bias when calibrating risk thresholds.